Machine Learning Architecture for Combating Hate Speech in Igbo Language

This research embarked upon the creation and integration of an Igbo Language Hate Speech Detection System into the Facebook platform. It commenced with a deep dive into the landscape of hate speech and its nuances in Igbo, the study identified a critical need for language-specific solutions. The implementation phase involved rigorous data preprocessing, focusing on cleaning the dataset by removing non-English characters and special symbols. Tokenization and lemmatization techniques were then applied to enhance the quality of the textual data. The TF-IDF vectorization method played a pivotal role in converting text into numerical features, laying the groundwork for a robust neural network model designed using Keras. Extensive testing, including precision, recall, and F1-score assessments, underscored the model's effectiveness in distinguishing hate speech from non-hate speech. Visualizations, such as word clouds and heatmaps, provided valuable insights into the dataset and model training process. The research also emphasized the user-centric approach with the integration of a user-friendly graphical interface using Kivy. This interface, featuring the Montserrat-Medium font and error handling mechanisms, not only enhances accessibility but also contributes to the overall reliability of the system. Regular monitoring, model retraining to accommodate evolving language patterns, security audits, user feedback analysis, and documentation updates form the pillars of a comprehensive maintenance plan. This ensures the sustained performance, security, and user satisfaction of the Hate Speech Detection System on the Facebook platform.